is.na behavior change/regression on Windows + r-devel

[MichaelChirico]

Seen in recent runs e.g.

https://github.com/Rdatatable/data.table/actions/runs/13041566732

Error in test.data.table() : 
  61 errors out of 11756. Search tests/tests.Rraw for test numbers 894.00113, 894.00123, 894.00133, 894.00143, 894.00153, 900.2, 901.1, 901.2, 908, 1010.4, 1010.6, 1267.1, 1267.2, 1336.1, 1336.2, 1343.3, 1343.4, 1343.5, 1343.6, 1344.2, 1344.4, 1344.6, 1550.1, 1550.2, 1550.3, 1550.4, 1550.5, 1550.6, 1550.7, 1550.8, 1550.9, 1551.61, 1551.62, 1552.1, 1552.2, 1552.3, 1552.4, 1552.9, 1555.12, 1635.1, 1751.4, 1753.2, 1755, 1768, 1825.04, 1827.1, 1827.3, 1840.2, 1855.2, 1855.3, 1857.02, 1857.03, 1867.02, 1867.03, 1867.08, 1867.09, 1870.3, 1870.4, 1870.5, 1883, 1903.2. Duration: 44.0s elapsed (35.3s cpu).

I'm not sure the cause, but filing first. It could well just be an r-devel bug that will resolve itself.

[MichaelChirico]

I don't reproduce the issue locally on r87668 on Linux. The failing run is r87664. The last passing test we have in Windows GHA is r87642.

[MichaelChirico]

I also don't reproduce on r87664 locally. But we do see the error on Linux on CRAN (hence removed the platform-specific label).

[MichaelChirico]

Locally I have:

gcc (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0

The failing CRAN server is using:

GCC 14.2.0 (Debian 14-2.0-12)

GHA has

* R was compiled by
    gcc.exe (GCC) 13.3.0
    GNU Fortran (GCC) 13.3.0

Maybe my gcc is too old...

[MichaelChirico]

Duplicate of #6779, which has more useful info